SIAgentNews.com
← All news

Microsoft launches specialized AI models to lower costs and boost productivity

The tech giant is moving away from 'one-size-fits-all' AI, releasing targeted models for coding, security, and office tasks at significantly lower prices.

· First reported by Zee Business on MSN

5%Price of MAI Code 2 vs. Opus 5.5

The short version

  • MAI Code 2 reaches 96% of the performance of Anthropic's Opus 5.5 at only 5% of the price.
  • The new MAI Cyber 2 Flash model beats GPT-6 Astra's quality at less than 2% of the cost per task.
  • Dragon Copilot Transcribe powers 10 million patient encounters a month with halved multilingual error rates.
  • Microsoft's new MAI Work Flash tool helps with Excel formulas, Word collaboration, and Teams summaries.

Why it matters

For businesses, this shift means they can potentially reduce the high costs of running AI while getting better results for specific tasks. It signals a move toward 'purpose-built' software where tools are optimized for your specific job rather than general conversation.

A shift toward specialized AI

Microsoft recently revealed a new series of artificial intelligence models during its Ignite 2026 conference in San Francisco. Instead of creating one massive model to do everything, the company is focusing on building specialized tools for distinct tasks like coding, cybersecurity, and office productivity. This strategy aims to provide high-quality features at a much lower price point than current market leaders.

The company's AI chief, Mustafa Suleyman, explained that this approach helps balance performance with the high costs of running these systems. By creating models for particular uses, Microsoft hopes to make AI more accessible for different types of work. This move is a direct response to the fierce competition from other major developers like OpenAI and Anthropic.

These new models are being marketed as a way to stand out in a crowded market. While many users are used to general-purpose chatbots, Microsoft is betting that businesses want tools that are specifically tuned for their unique workflows. This transition focuses on efficiency and cost-effectiveness for large-scale operations.

Coding and cybersecurity tools

One of the major highlights is MAI Code 2, a model designed to assist with computer programming. Microsoft claims it reaches 96% of the performance of Anthropic's Opus 5.5 but costs only 5% of the price. For developers who need a lighter solution for everyday tasks, the company also released MAI Code 2 Flash, which matches the performance of other low-cost models while remaining more affordable.

In the realm of security, Microsoft introduced a dedicated AI model to identify and fix software vulnerabilities. This comes as it becomes increasingly easy for people to use AI tools to find flaws in systems. The company plans to roll this out first to a select group of customers to help organizations protect themselves before they can be exploited by bad actors.

The company also highlighted MAI Cyber 2 Flash, which reportedly beats GPT-6 Astra’s quality at less than 2% of the cost per task on a specific benchmark. These tools are designed to help organizations manage the risks associated with AI safety. Suleyman noted the importance of ensuring these systems do not engage in harmful actions and instead follow their intended purposes.

Boosting office productivity

For the average office worker, Microsoft introduced MAI Work Flash. This tool is not intended to be a coding assistant but rather a helper for common daily tasks. It is built on the Code 2 Flash architecture and is designed to integrate with the tools people already use every day, such as Excel, Word, and Microsoft Teams.

Users can use these new features to build complex formulas in Excel or collaborate on documents in Word. The tool can also help summarize and extract key information from conversations held within Microsoft Teams. This makes the AI more practical for people who need to get work done quickly rather than just seeking general information from a chatbot.

Additionally, the company updated its image generation and speech transcription capabilities. Dragon Copilot Transcribe, built on the MAI Transcribe 2 model, is already being used for 10 million patient encounters a month. This model is described as the world’s fastest and most accurate multilingual transcription tool, featuring halved error rates for different languages.

Hardware and cost efficiency

To support these new models, Microsoft is also improving its underlying hardware. The company announced Maia 200, its second-generation AI accelerator, which is now in production. This hardware is designed to run Microsoft's largest models while achieving 1.7x more tokens per watt, which refers to the amount of data processed per unit of energy used.

Efficiency is a core theme of this announcement, as evidenced by the Decision 1.1 model. This new decision model is reported to be 50x faster and 79x cheaper than the GPT-Sol 6.1 model. By optimizing both the software models and the hardware they run on, Microsoft aims to lower the barrier for companies to adopt AI at scale.

However, Microsoft clarified that these performance figures are tied to API usage costs. The company noted that these numbers are not intended to suggest that users can simply switch their current monthly subscriptions for a lower cost. Instead, the focus is on providing a more efficient infrastructure for developers and large organizations who pay based on how much they use the service.

What to watch

  • How independent tests will measure the actual performance of MAI Code 2 in real-world coding tasks.
  • Which specific customers will be the first to receive the new cybersecurity AI model.

Try it

Check if your workplace uses Microsoft Teams or Excel to see if you can start using the new MAI Work Flash features for formula building or conversation summaries.

Sources and links

First reported by Zee Business on MSN. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

More AI news